This paper proposes the Intuitive Systems framework, a reinforcement learning (RL)-based methodological approach for generative architectural design. Rather than treating RL as a post-optimization tool, the framework restructures the generative process into a two-stage workflow: a policy-training phase driven by generalized design criteria, and a case-specific formation phase in which the trained policy guides spatial generation. Through this structure, generative systems are cultivated as adaptive agents capable of developing reusable design strategies across varying environmental conditions. Two design experiments—Domino Walk and Intuitive Field—demonstrate how RL can mediate between stochastic form exploration and performance-based constraints. The results show that embedding quantitative design intentions within reward structures enables controlled variability while maintaining responsiveness to spatial, structural, and environmental criteria. Compared to evolutionary selection or data-driven generative models, the proposed approach emphasizes strategy formation through iterative interaction, enabling adaptable behavioral logic rather than instance-based optimization. By articulating the operational relationship between reinforcement learning mechanisms and architectural generative logic, this research addresses a methodological gap between machine learning frameworks and design reasoning. The Intuitive Systems framework contributes a structured pathway for integrating adaptive computational intelligence into architectural formation, supporting scalable, controllable, and context-aware generative design processes.
Wang et al. (Mon,) studied this question.